NHS keeps Palantir until 2029 and skips its own break clause

Britain had a contractual right to walk away from Palantir's NHS data platform next spring. On Monday it chose two more years instead — and set a replacement deadline its own procurement calendar can barely hit.
The break clause that wasn't used
Health minister James Frith told MPs on Monday that the Federated Data Platform (FDP) contract — the software layer that links patient data across NHS trusts in England — will run until March 2029 rather than ending at the break point next spring, the BBC reported. The deal was awarded in November 2023 to a Palantir-led group, has a maximum value of £330 million and includes other suppliers. Its seven-year term was built as a three-year initial period ending in March 2027, with extension options of two years, then one, then one. Ministers have now taken the two-year option and said the contract ends there.
The replacement timetable is the part to read twice. Formal procurement for a "new FDP iteration" begins in the second half of 2027 and is expected to take around 18 months, Frith said. The new platform is meant to be built around interoperability, resilience and common standards, with the first new capability pointed at maternity services; bidders will be welcome from UK small firms, "sovereign AI-backed capability" and international providers alike. Frith called the extension "a bridge, not the destination," Politico reported.
Two parliamentary committees had called for the contract to be ended this year. Labour MPs who wanted the break clause used called the deal "an albatross around the NHS's neck"; the Liberal Democrats called it dangerous; the BMA said it was glad ministers had listened but frustrated by the delay. Campaigners object to Palantir's work for the Israeli military and for US immigration enforcement. Frith's answer was that the decision is "not about the politics of executives of suppliers."
The evidence under the platform is thinner than the contract
The awkward fact for the government is that the case for the FDP's performance rests largely on claims the government has had to withdraw. Last week NHS England deleted a line from its website saying 137 hospital trusts were benefiting from Palantir's system, after the Guardian reported that 29 of those trusts said the platform was either not live or that they were using none of its applications. The Office for Statistics Regulation is examining the claims, and had already warned NHS England that it had no evidence for a claim of more than 100,000 extra operations attributed to the platform.
That pattern — a benefit number published, then quietly removed — is the same failure mode we described when AI agents flood public services — and the claims are mostly real: the tools often work, but the reporting around them is written by the people selling them. Palantir's own record with the NHS adds context. It collected around £60 million in uncontested NHS contracts during and after the pandemic, and those use cases were carried into the FDP procurement. NHS England has since asked for a second opinion on Palantir's self-assessments, and the company's NHS cancer app has drawn praise alongside reports of very few users. The government's answer to all of this was to give the vendor two more years.
Palantir's defence, relayed by its UK senior vice president Tom Watson — the former Labour deputy leader, now a full-time company executive — is that the programme is "significantly exceeding its benefits forecast and has exceeded every target" since 2024, in ministers' words. Watson said the firm will "deepen our investment in the UK" over the next two years. That is a vendor promising to embed itself harder in the thing it is being asked to leave.
"It is not a break clause; it is a lock-in"
Dame Chi Onwurah, who chairs the Commons science and technology committee, put the central question to Frith directly: if a break clause cannot be exercised because of continuity of care, what was it for? "If a break clause cannot be exercised because of continuity of care, it is not a break clause; it is a lock-in," she told the Commons, per The Register. Frith confirmed the contract contains exit and transition provisions and that migration was "possible in principle", but argued that 18 to 24 months for commissioning and procurement was a reasonable assessment.
Both statements can be true. Migrating a national data platform that hospitals use to manage waiting lists is genuinely risky, and pretending otherwise is how health IT projects fail. But the risk is manufactured by the calendar as much as by the software: a government that intended to leave could have started the tender in 2025 instead of the second half of 2027, which is why campaigners keep pointing at the contract's "termination assistance period" as a cheaper, faster exit than the one ministers chose. The 2029 date is not a technical constraint. It is a scheduling decision dressed as one.
There is a second, quieter reason this matters beyond procurement law. The FDP is the data layer under every AI tool the NHS wants to run across trust-level patient records — triage, discharge prediction, imaging queues, theatre scheduling. Whoever operates that layer decides what models can be pointed at it, under what governance, and with whose audit trail. A two-year extension is not just two more years of licence fees; it is two more years in which the substrate for English clinical AI belongs to one American company, while the minister talks about sovereign capability. Frith would not say whether Palantir can bid for the replacement contract. That silence is the most revealing sentence of the announcement.
Who wins, who loses
Palantir wins the revenue and, more importantly, the embedding: another 28 months to add modules, integrations and the switching costs that make the next exit harder than this one. It also keeps growing elsewhere in Britain — three police forces signed a £10.1 million deal on 13 November — even as one police unit pulled the plug on a Palantir AI tool, which tells you the UK market is not closing.
UK tech firms get a genuine invitation and no guarantee. The new tender's emphasis on interoperability favours whoever already runs the integration layer, and the procurement will be judged on delivery risk by the same officials who just argued that switching is too risky to attempt. The losers, if the benefits claims are as shaky as the regulator suggests, are the trusts paying for shelfware and the patients whose waiting lists were used to sell it.
The one thing that would settle the argument is independent measurement, which is also the bet behind OpenAI funds $5M toolkit for government AI watchdogs. Britain has a statistics regulator willing to ask the question. It does not yet have a procurement process willing to act on the answer.
What to watch
The Office for Statistics Regulation's verdict on NHS England's benefit claims, due before the new tender is drafted. Whether the 2029 procurement is written as open standards with portable data or as a re-badged FDP that only the incumbent can credibly deliver. And whether the maternity-services pilot, the first declared piece of the "new iteration", goes to a UK supplier or becomes the proof point for keeping the current one. If the answer to all three is "the incumbent", the bridge will turn out to have been the destination after all.
Should a public health service be allowed to renew a disputed data contract on continuity grounds, or should the break clause be a real exit? Write to us via our contact page.







